Model the dynamics, or robustify the standard errors?

Question

Your time-series regression starts with autocorrelated residuals. You add a couple of lagged regressors (and a lagged dependent variable), and now the residuals pass a white-noise test cleanly.

Do you still need Newey-West HAC standard errors? Compare the two philosophies: modeling the dynamics versus robustifying the standard errors.

Accepted Answer

If the residuals are genuinely white noise, you do not need HAC for autocorrelation. HAC standard errors exist to correct inference when residuals are serially correlated. If adding lags has removed that correlation, the residuals now satisfy the no-autocorrelation assumption, and ordinary or White standard errors are valid. (You may still want White standard errors if heteroskedasticity remains.) The two philosophies. - Model the dynamics. Include enough lagged terms to soak up the serial dependence, so the errors become white noise. Upside: if the dynamics are of genuine interest or improve…

Model the dynamics, or robustify the standard errors?

Your answer

Discussion

Learn the concepts

Related questions